Early warning system and early warning method for monitoring ward based on Internet of Things

Abstract

The invention belongs to the technical field of early warning systems in Internet platforms, and discloses an Internet of Things-based early warning system and early warning method for an intensive care unit, and the system carries out the abnormal recognition of physiological parameters through a pre-trained early warning model, fuses response task features, medical worker resource states and historical scores based on a response scheduling module, and carries out the early warning of the medical worker. A responder assignment strategy is optimized, and the matching accuracy is improved; furthermore, periodic data acquisition is realized through a response closed-loop recording module, an early warning response responsibility chain containing node and causal link information is constructed, and on the basis, high-risk nodes and bottleneck regions are identified, task allocation weights and path optimization rules are dynamically adjusted, and adaptive updating of scheduling logic is realized; and early warning re-response and rescheduling based on new logic are supported, and a feedback closed loop driven by data is formed, so that the timeliness and accuracy of early warning response are effectively improved, and the risk of intervention failure caused by response lag is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of early warning systems in Internet platforms, and in particular relates to an early warning system and an early warning method for an intensive care unit based on the Internet of Things. Background Art

[0002] With the widespread application of IoT technology in healthcare settings, IoT-based ICU management has gradually enabled real-time collection and remote monitoring of patients' vital signs. Existing technologies typically deploy multi-parameter monitoring devices to collect physiological parameters such as heart rate, blood pressure, and blood oxygen levels. Alarms are triggered based on fixed threshold rules, prompting medical staff to intervene. Some systems also incorporate simple task dispatching and rotation scheduling mechanisms to improve response efficiency. However, several issues remain in practice: First, current early warning mechanisms have limited ability to identify abnormal conditions, prone to missed or false alarms; second, most response processes fail to consider the real-time status and workload of medical staff, resulting in low response efficiency; and third, a lack of structured documentation of the entire early warning response process makes it impossible to identify process bottlenecks. These issues severely limit the intelligence level of early warning management and the closed-loop response capability, and urgently need improvement. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an early warning system and early warning method for intensive care units based on the Internet of Things, so as to solve the technical problems of low early warning accuracy, low response efficiency and low intelligence level in the early warning management of intensive care units in the existing technology.

[0004] The first aspect of the present invention discloses an early warning system for an intensive care unit based on the Internet of Things, which includes a vital sign collection module, an early warning identification module, a response scheduling module, a response closed-loop recording module, and an early warning feedback module; wherein,

[0005] Vital signs collection module, used to collect patients' physiological parameter data in real time;

[0006] an early warning identification module, configured to perform an abnormal state identification operation based on the physiological parameter data using a pre-trained early warning model and output a first early warning message;

[0007] a response scheduling module, configured to determine a response task based on the first warning information, determine a response personnel based on the response task, warning generation unit information, the current location of the medical staff, task load, professional matching, and historical response efficiency score, and assign the response task to the response personnel;

[0008] The response closed-loop recording module is used to record the response arrival status, intervention operation type and processing results of medical staff according to a preset period as response closed-loop data;

[0009] The early warning feedback module is used to build an early warning response responsibility chain based on the first early warning information within the cycle and the response closed-loop data, and perform scheduling logic adjustment operations based on the early warning response responsibility chain.

[0010] Furthermore, the warning response responsibility chain includes node information and link relationships; wherein,

[0011] The node information includes the early warning generation unit identifier, the dispatching unit identifier, the responder identifier, the intervention operation type and the intervention result status;

[0012] The link relationship includes the delay relationship, task dependency and operation causal relationship between nodes.

[0013] Furthermore, the process of constructing the early warning response responsibility chain includes:

[0014] Conduct temporal correlation analysis on each warning event and reconstruct the response path from warning generation, task dispatch, response arrival, intervention execution to processing completion;

[0015] Through the graph modeling mechanism, each response link is abstracted as a node according to the response path, and the operation timing, data dependency and causal relationship are mapped into edge relationships, generating a warning response responsibility chain structure diagram with response path, multi-stage delay parameters between nodes, intervention action information and task result identification.

[0016] Furthermore, after constructing the early warning response responsibility chain structure diagram, the high-risk nodes and response bottleneck areas in the early warning process are identified based on the early warning response responsibility chain structure diagram. The high-risk node identification process includes:

[0017] A multi-dimensional performance evaluation index is constructed based on the average response delay, task completion rate and intervention effectiveness of each node in the early warning response responsibility chain structure diagram;

[0018] A weighted calculation is performed on each node according to the multi-dimensional performance evaluation indicators to obtain the bottleneck risk score of each node. Nodes with bottleneck risk scores higher than the preset threshold are regarded as high-risk nodes.

[0019] Furthermore, the process of identifying the response bottleneck area in the early warning process includes:

[0020] Based on the adjacency relationship and path overlap of high-risk nodes in the early warning response responsibility chain structure diagram, areas with continuous high-risk nodes or high overlap of multiple early warning task paths are identified as response bottleneck areas.

[0021] Furthermore, the execution of the scheduling logic adjustment operation specifically includes:

[0022] Dynamically adjust the priority weights of medical staff's task allocation based on the type and spatial distribution of high-risk nodes;

[0023] Adjust the path optimization rules in the early warning model based on bottleneck area feedback;

[0024] Determine the response frequency and path coverage corresponding to the response bottleneck area, and update the load distribution constraints in the task scheduling process based on the response frequency and path coverage.

[0025] Furthermore, the first warning information includes the abnormality type, abnormality occurrence time, confidence score and warning generation unit information;

[0026] After executing the scheduling logic adjustment operation, the warning identification module and the response scheduling module are further used to perform a warning response based on the adjusted scheduling logic; wherein,

[0027] The warning identification module performs a warning response based on the adjusted scheduling logic including:

[0028] The warning model generates second warning information for the newly acquired physiological parameter data based on the adjusted path optimization rule; the warning generation unit information in the second warning information is determined based on the adjusted path optimization rule.

[0029] Furthermore, the response scheduling module performs an early warning response based on the adjusted scheduling logic, including:

[0030] When the number of determined responding personnel is lower than the preset minimum selectable personnel threshold, the current task scheduling resources are adjusted according to the updated load distribution constraints;

[0031] When the number of identified responders is more than one, the target responders are determined based on the adjusted priority weights of the medical staff's tasks.

[0032] Furthermore, the response scheduling module is also used to monitor the execution status of the response task. If the responder fails to complete the response within the specified time, the backup responder redistribution operation is triggered.

[0033] A second aspect of the present invention discloses an early warning method for an intensive care unit based on the Internet of Things, which is applied to the system disclosed in the first aspect and includes:

[0034] Collect the patient's physiological parameter data in real time;

[0035] Performing an abnormal state recognition operation based on the physiological parameter data through a pre-trained early warning model and outputting a first early warning message;

[0036] Determine the response task based on the first warning information, and determine the responder based on the response task, warning generation unit information, the current location of the medical staff, task load, professional matching and historical response efficiency score, and assign the response task to the responder;

[0037] Record the medical staff's response arrival, intervention type, and treatment results according to the preset cycle as response closed-loop data;

[0038] An early warning response responsibility chain is constructed based on the first early warning information within the cycle and the response closed-loop data, and a scheduling logic adjustment operation is performed based on the early warning response responsibility chain.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention realizes intelligent response and continuous optimization of the early warning process through the collaboration of multiple modules. First, a pre-trained early warning model is used to identify abnormal physiological parameters. Based on the response scheduling module, the response task characteristics, the resource status of medical staff and the historical scores are integrated to optimize the response personnel assignment strategy and improve the matching accuracy. Furthermore, periodic data collection is realized through the response closed-loop recording module, and an early warning response responsibility chain containing node and causal link information is constructed. On this basis, high-risk nodes and bottleneck areas are identified, and the task allocation weights and path optimization rules are dynamically adjusted to realize the adaptive update of the scheduling logic. It also supports early warning re-response and re-scheduling based on the new logic, forming a data-driven feedback closed loop, thereby effectively improving the timeliness and accuracy of the early warning response and reducing the risk of intervention failure caused by delayed response. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0042] Figure 1 This is a structural diagram of an early warning system for an intensive care unit based on the Internet of Things disclosed in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0044] Example 1 The first aspect of the present invention discloses an early warning system for intensive care units based on the Internet of Things. Figure 1 , Figure 1This is a schematic diagram of the structure of an early warning system for an intensive care unit based on the Internet of Things disclosed in an embodiment of the present invention. The system includes a vital sign collection module, an early warning identification module, a response scheduling module, a response closed-loop recording module, and an early warning feedback module; wherein,

[0045] Vital signs collection module, used to collect patients' physiological parameter data in real time;

[0046] an early warning identification module, configured to perform an abnormal state identification operation based on the physiological parameter data using a pre-trained early warning model and output a first early warning message;

[0047] a response scheduling module, configured to determine a response task based on the first warning information, determine a response personnel based on the response task, warning generation unit information, the current location of the medical staff, task load, professional matching, and historical response efficiency score, and assign the response task to the response personnel;

[0048] The response closed-loop recording module is used to record the response arrival status, intervention operation type and processing results of medical staff according to a preset period as response closed-loop data;

[0049] The early warning feedback module is used to build an early warning response responsibility chain based on the first early warning information within the cycle and the response closed-loop data, and perform scheduling logic adjustment operations based on the early warning response responsibility chain.

[0050] It can be understood that in the actual deployment process, the various functional modules in the present invention are integrated into a unified platform through the Internet of Things architecture. Specifically, the vital signs acquisition module establishes a communication connection with the hospital local area network or edge server through an embedded communication module (such as a Wi-Fi or NB-IoT module) to achieve quasi-real-time uploading of patient status data. The early warning identification module is deployed on the edge computing node or cloud server, and performs abnormal status identification on the continuously collected data. The identification results form early warning information and are transmitted to the response scheduling module through the network. The response scheduling module obtains the location information and current task status of the medical staff based on the smart terminals worn by the medical staff (such as positioning badges and wristbands), determines the target response personnel in real time through the scheduling algorithm, and issues task information. After the task is issued, the response closed-loop recording module recovers various response status data through the Internet of Things communication protocol (such as MQTT) for subsequent model feedback and scheduling logic optimization.

[0051] Specifically, in an embodiment of the present invention, a vital signs acquisition module is deployed in a bedside environment in an intensive care unit, including but not limited to a variety of medical-grade physiological parameter sensors for real-time collection of core vital signs data of patients. The collected data includes at least electrocardiogram (ECG) signals, blood pressure and pulse waves, blood oxygen saturation, body temperature and respiratory rate, consciousness score and abnormal behavior indicators, etc. Furthermore, ECG signals are continuously monitored through multi-lead electrode patches to identify abnormal conditions such as arrhythmias; blood pressure and pulse waves are dynamically recorded using a non-invasive continuous blood pressure module and a PPG sensor; blood oxygen saturation is obtained through a finger-clip photoelectric probe; body temperature and respiratory rate: measured using a surface temperature sensor and a chest strain belt; consciousness score and abnormal behavior indicators can be obtained through a pressure sensor under the mattress or visual recognition to obtain the patient's body movement frequency and call frequency.

[0052] This sensor data is locally aggregated via edge acquisition terminals and regularly uploaded to the server using low-power Bluetooth or Wi-Fi modules. Each uploaded data packet carries a unique patient ID, timestamp, and sensor source ID to ensure data traceability and continuity.

[0053] During the early warning recognition process, the early warning recognition module performs periodic anomaly detection and status identification on the aforementioned physiological parameter data based on a pre-trained AI model deployed on edge computing nodes. This early warning model is a deep time series model with path scoring perception capabilities. Its reasoning process not only outputs anomaly results but also embeds a path optimization mechanism to generate warning generation units within the early warning response path.

[0054] Specifically, the model includes a temporal feature extraction layer, an anomaly scoring and classification unit, and a path optimization scoring unit. In the temporal feature extraction layer, a multi-channel GRU and convolutional unit fusion architecture is used to extract features from both the time dimension and the short-term fluctuation dimension. In the anomaly scoring and classification unit, a multi-class anomaly identifier based on cluster center offset and self-attention mechanism is constructed to output anomaly type, occurrence time, and anomaly confidence score. The path optimization scoring unit embeds a set of scoring functions in the model output stage, combines the current ward layout and response scheduling history, calculates the path superiority score of each candidate warning generation unit in each available path, and ultimately selects the optimal generation path and node to form structured warning generation unit information.

[0055] Based on the above structure, the first warning information is the comprehensive output of the model, with a format including but not limited to the anomaly type, anomaly occurrence time, anomaly confidence score, and warning generation unit ID. The warning generation unit is determined by the path optimization scoring unit based on the ward network structure and timeliness scoring function. For example, the edge node with local transit dispatch authority and farthest from the response bottleneck is selected as the generation source to improve overall response efficiency.

[0056] The early warning model training process uses a dataset constructed from historical data from real ICUs. The samples include normal segments, physiological abnormalities, medical response records, and closed-loop event information. Labels are annotated by experts and aligned with closed-loop feedback. A weighted multi-objective loss function is used during training, with objectives including anomaly identification accuracy, response path score consistency, and response delay prediction error. This ensures that the model not only accurately identifies anomalies but also generates the optimal early warning path.

[0057] Through the above-mentioned settings, the early warning model of the present invention has the ability to integrate path output and scoring, embedding path logic into the model structure, ensuring that the early warning information has responsive directionality, and avoiding the structural separation problem of traditional path generation that relies on the post-scheduling system. In addition, the vital signs collection module and the early warning identification module together constitute the system's perception layer and intelligent analysis front end. Through the continuous collection and efficient modeling of patient status, it provides high-confidence abnormality identification results, and combines the path optimization strategy to generate structured early warning information, providing a precise and reliable execution basis for the subsequent scheduling module.

[0058] Furthermore, the early warning response responsibility chain includes node information and link relationships;

[0059] The node information includes the early warning generation unit identifier, the dispatching unit identifier, the responder identifier, the intervention operation type and the intervention result status;

[0060] The link relationship includes the delay relationship, task dependency and operation causal relationship between nodes.

[0061] Furthermore, the process of building the early warning response responsibility chain includes:

[0062] Conduct temporal correlation analysis on each warning event and reconstruct the response path from warning generation, task dispatch, response arrival, intervention execution to processing completion;

[0063] Through the graph modeling mechanism, each response link is abstracted as a node according to the response path, and the operation timing, data dependency and causal relationship are mapped into edge relationships, generating a warning response responsibility chain structure diagram with response path, multi-stage delay parameters between nodes, intervention action information and task result identification.

[0064] Specifically, to fully track the early warning process and identify bottlenecks in the ICU, this paper introduces an early warning response responsibility chain to record, analyze, and dynamically provide feedback on the execution status of each step in the early warning event processing. This responsibility chain, using a graph structure, reflects the entire execution process from early warning generation to closed-loop response.

[0065] The early warning response responsibility chain consists of two basic elements: node information and link relationships. Specifically, each early warning response responsibility chain contains multiple nodes, each representing a specific responsible entity or operational link in the early warning process. The early warning generation unit identifier indicates which early warning unit in the system generated the early warning, such as a specific ward terminal device. The dispatch unit identifier represents the system unit that performs early warning information scheduling and task distribution operations, such as a local server, scheduling platform, or edge scheduling agent. The responder identifier refers to the identity code of the medical staff who clearly received the task and entered the response state. The intervention operation type indicates the type of intervention measure performed by the recorded responder, such as medical operation identifiers such as oxygen inhalation and cardioversion. The intervention result status reflects the completion status and effectiveness of the intervention, such as success, partial completion, failure, or requiring secondary processing. This structure gives each node clear behavioral semantics, providing a quantifiable basis for subsequent node performance evaluation and responsibility attribution.

[0066] Link relationships represent the logical and temporal relationships between nodes. Latency relationships refer to the response delay between two adjacent nodes, including multi-stage delays such as the dispatch delay from warning task generation to dispatch, the response delay from dispatch to response reception, the intervention decision delay from response arrival to intervention, and the intervention disposition delay from intervention to completion. Task dependency relationships indicate whether the current node in a record depends on the completion of the previous node. For example, a secondary measurement of physiological parameters can only be performed after a drug injection is completed. Operational causality indicates whether an operation is directly triggered by a previous operation.

[0067] In addition, in an embodiment of the present invention, the generation process of the early warning response responsibility chain is based on the full-process data tracking and behavior tracking of each early warning event. First, through the temporal analysis of the early warning related data, the early warning generation records, task dispatch records, response personnel receiving information, on-site arrival time, intervention behavior execution status and final result status collected within a certain early warning cycle are time-aligned one by one. Through the sliding time window and filtering algorithm, redundant trigger records are eliminated to ensure that each early warning event is uniquely and completely mapped to a clear response path. Then, based on the behavioral semantics, the key links in the early warning process are abstracted as nodes in the graph, including but not limited to early warning generation nodes, dispatch nodes, response receiving nodes, intervention execution nodes and processing completion nodes, etc. Each node is bound to the corresponding execution unit information, operation type and processing status.

[0068] During the construction process, the system automatically extracts the time intervals between nodes, recording the actual response delays at each stage, from alert generation to response processing. These delay parameters include both task scheduling intervals and the actual time required for on-site arrival and intervention execution, providing a reliable foundation for subsequent performance evaluation and bottleneck analysis. Furthermore, the system appends the intervention type and execution result in each response path to the corresponding node, serving as task label information in the responsibility chain diagram to reflect the actual operational behavior and intervention effect of each response task.

[0069] Through the graph modeling mechanism, these nodes are connected into a chain in chronological order, and multi-dimensional edge relationships are formed based on the time intervals, data dependencies and causal logic between nodes to form a preliminary responsibility chain graph structure.

[0070] As a preferred method, on the basis of constructing the preliminary early warning response responsibility chain structure diagram, the structure diagram is further enhanced. First, the metadata bound to each node is expanded. On the basis of the original execution unit identification, operation type and processing status, statistical attributes related to the historical behavior of the node are further introduced, such as average response time, intervention success rate, adjacent node jump frequency, etc.; in addition, to support path-level analysis, path labels, operation priority identification and location distribution parameters are added to some key nodes to make them more expressive. In the enhancement of edge relationships, the temporal relationship, task dependency and causal logic between nodes are retained, and weighted graph edge weights are constructed, in which the edge weight comprehensively considers the path frequency, response impact and sensitivity parameters under critical time periods. For example, for the edge relationship between two consecutive response links, if historical data shows that the segment often has delayed or erroneous responses, the edge weight risk value is increased for subsequent bottleneck detection.

[0071] Furthermore, for warning response paths with repeated structures in multiple cycles, frequently occurring high-reuse task segments are abstracted into path cluster nodes through sub-path clustering and graph compression mechanism, and their behavioral characteristics are aggregated and statistically analyzed, which improves the graph compression rate while enhancing the correlation representation between tasks.

[0072] Through this process, a complete responsibility chain structure diagram is generated, depicting the entire process of each warning event, from its transfer, execution, and closure. This structure diagram not only preserves the behavioral semantics and status labels of task nodes, but also fully describes the logical sequence of task flows and the temporal characteristics of response links. This provides a unified data foundation and structural support for subsequent high-risk and bottleneck identification, scheduling optimization, and model feedback.

[0073] Furthermore, after constructing the early warning response responsibility chain structure diagram, the high-risk nodes and response bottleneck areas in the early warning process are identified based on the early warning response responsibility chain structure diagram. The high-risk node identification process includes:

[0074] A multi-dimensional performance evaluation index is constructed based on the average response delay, task completion rate and intervention effectiveness of each node in the early warning response responsibility chain structure diagram;

[0075] A weighted calculation is performed on each node according to the multi-dimensional performance evaluation indicators to obtain the bottleneck risk score of each node. Nodes with bottleneck risk scores higher than the preset threshold are regarded as high-risk nodes.

[0076] Furthermore, the process of identifying the response bottleneck area in the early warning process includes:

[0077] Based on the adjacency relationship and path overlap of high-risk nodes in the early warning response responsibility chain structure diagram, areas with continuous high-risk nodes or high overlap of multiple early warning task paths are identified as response bottleneck areas.

[0078] Specifically, in an embodiment of the present invention, after constructing the early warning response responsibility chain structure diagram, a fine-grained analysis is further performed on each response node in the responsibility chain to identify key links where response delays, intervention failures, or task omissions repeatedly occur during multiple early warning response processes, thereby forming a basis for determining high-risk nodes and response bottleneck areas.

[0079] To accurately identify high-risk nodes, the present invention first extracts the basic performance indicators of all nodes from the responsibility chain structure diagram, including but not limited to the average response delay, task completion rate and intervention effectiveness (such as the probability of patient status stabilization or symptom improvement ratio) corresponding to the node in multiple early warning tasks.

[0080] In the above three dimensions, the average response delay can be obtained by calculating the average of the stage delays of the node in historical tasks; the task completion rate is the ratio of the number of successful completions of the node's duties and tasks to the total number of node participations; and the effectiveness of the intervention can be combined with the changing trends of the patient's physiological parameters before and after the intervention, supplemented by symptom records and medical feedback, to quantify the intervention score within a certain range. The three types of indicators are constructed into a standardized set of performance evaluation indicators, and the bottleneck risk score of each node is calculated by linear weighting. As a preferred method, in order to improve the accuracy of risk identification, each evaluation indicator is differentially weighted by experience weight, task priority or regional risk level to adapt to the differences in response characteristics under different wards or patient types.

[0081] When the bottleneck risk score of a node is continuously higher than the set warning threshold, it is marked as a high-risk node and subsequently analyzed. It should be noted that the high-risk node is not necessarily a specific medical individual, but may also be a structured node representing a certain type of task, a certain spatial area, or a certain intervention method. After identifying a group of high-risk nodes, their topological distribution in the responsibility chain diagram is further analyzed, and the potential response bottleneck area is identified by combining the node adjacency relationship and path overlap rate.

[0082] In this embodiment of the present invention, the logic for identifying response bottleneck areas is based on two core characteristics: first, the continuous distribution of high-risk nodes, i.e., multiple high-risk nodes are interconnected and influence each other in the same response path or adjacent paths; second, the path overlap rate, i.e., multiple independent warning tasks frequently share the same response path or the same type of key nodes during execution. By analyzing the overlap of paths in the responsibility chain, link areas that are frequently accessed or repeatedly carried are identified. Combined with the high-risk node density of these areas, they are ultimately determined to be response bottleneck areas for the current cycle.

[0083] Through the above-mentioned identification of high-risk nodes and determination of response bottleneck areas, it is possible to intelligently attribute the weak links in the ICU's early warning response process, providing a data basis and directional guidance for subsequent scheduling optimization.

[0084] As a preferred implementation, in order to quantitatively evaluate the bottleneck risk of each response node, the following risk scoring formula is introduced:

[0085]

[0086] in, is the comprehensive bottleneck risk score of node i; is the average response delay of node i; The maximum response delay of all nodes in the current cycle; is the coefficient of variation of the average response delay of node i, which is used to measure the stability of the node's response performance; is the task completion rate of node i; The historical fluctuation index of task completion rate is used to reflect its long-term performance fluctuations; Score the effectiveness of the intervention behavior associated with node i; is the density factor of node i in the bottleneck area; 、 、 are the weight coefficients of the response delay index, completion rate index and intervention effectiveness index, and their sum is 1.

[0087] The above operation introduces stability adjustment items (coefficient of variation), historical fluctuation items and regional density adjustment items on the basis of the original risk score. In actual applications, it can more accurately identify key bottleneck nodes that have response lags, are unstable for a long time, and are in high-frequency areas, thereby improving the comprehensiveness and accuracy of high-risk node identification.

[0088] As another preferred embodiment, in order to more comprehensively evaluate the structural overlap and resource competition risk between different warning task paths, the path overlap rate calculation is set as follows:

[0089]

[0090] in, is the path overlap rate of the warning task path; is the total number of warning tasks in the current cycle; For path With path The number of intersections between nodes; is the total number of nodes in the path; For path The importance weight of the corresponding task; For path Historical trigger frequency in the current cycle; is the normalization coefficient to ensure that the overlap ratio is between 0 and 1.

[0091] In this implementation, not only the node overlap density at the structural level is considered, but also the task weight level and historical frequency factor are introduced, so as to more realistically reflect the concentration trend of bottleneck areas in actual workloads, and then be used to assist in identifying response bottleneck areas with highly shared resources and potential resource conflicts.

[0092] As another preferred embodiment, in a specific scenario, in order to enhance the ability to perceive node risk trends, a time-weighted adjustment is performed on the static scoring results, and the following time correction formula is introduced:

[0093]

[0094] in, is the final adjusted bottleneck risk score of node i; It is the time derivative of the risk score, which is used to reflect the growth trend of the risk value in the most recent period; It is a time sensitivity parameter that controls the impact of the score growth trend on the final score.

[0095] This implementation method is mainly aimed at dynamic high-risk node identification scenarios. Through the above operations, it can effectively identify those hidden nodes whose risk values ​​have increased sharply in the recent period and may enter a bottleneck state, thereby providing more forward-looking intervention suggestions for task scheduling.

[0096] Furthermore, performing the scheduling logic adjustment operation specifically includes:

[0097] Dynamically adjust the priority weights of medical staff's task allocation based on the type and spatial distribution of high-risk nodes;

[0098] Adjust the path optimization rules in the early warning model based on bottleneck area feedback;

[0099] Determine the response frequency and path coverage corresponding to the response bottleneck area, and update the load distribution constraints in the task scheduling process based on the response frequency and path coverage.

[0100] Furthermore, the first warning information includes the abnormality type, the abnormality occurrence time, the confidence score and the warning generation unit information;

[0101] After executing the scheduling logic adjustment operation, the warning identification module and the response scheduling module are further used to perform a warning response based on the adjusted scheduling logic; wherein,

[0102] The warning identification module performs a warning response based on the adjusted scheduling logic including:

[0103] The warning model generates second warning information for the newly acquired physiological parameter data based on the adjusted path optimization rule; the warning generation unit information in the second warning information is determined based on the adjusted path optimization rule.

[0104] Furthermore, the response scheduling module performs an early warning response based on the adjusted scheduling logic, including:

[0105] When the number of determined responding personnel is lower than the preset minimum selectable personnel threshold, the current task scheduling resources are adjusted according to the updated load distribution constraints;

[0106] When the number of identified responders is more than one, the target responders are determined based on the adjusted priority weights of the medical staff's tasks.

[0107] In this embodiment of the present invention, after constructing and analyzing the early warning response responsibility chain, the early warning feedback module dynamically adjusts the scheduling logic based on the identified high-risk node types and spatial distribution characteristics. This adjustment covers multiple aspects, aiming to optimize the dispatch strategy for subsequent early warning tasks, reduce response delays and the risk of intervention failure, and thus comprehensively enhance the intelligent response capabilities of the ICU.

[0108] Specifically, the priority weights of medical personnel's task assignments are dynamically adjusted based on the node types corresponding to high-risk nodes (such as nodes with frequently delayed response arrivals or intervention nodes with high intervention failure rates) and their spatial clustering. Priority weights constitute a key scheduling factor in task assignment and are weighted based on factors such as the responder's historical performance, specialty suitability, and current workload. When multiple high-risk nodes exist within a given area, priority is given to dispatching medical resources with a higher historical response efficiency to that area, thereby improving the local area's task processing capabilities.

[0109] Secondly, the bottleneck analysis results are used as feedback to update the path optimization rules in the early warning identification module. The path optimization rules refer to the scoring mechanism used to determine the warning generation unit or triggering path during the early warning model inference process. This multi-factor decision is made by combining expected response delay, overload probability, and historical bottleneck distribution. By strengthening the penalty coefficient for identified bottleneck paths or optimizing the scoring factor for unobstructed paths, the reuse of highly congested paths is effectively avoided, enhancing the adaptability of the early warning model to resource-constrained areas.

[0110] Furthermore, the present invention performs statistical analysis on the response frequency and path coverage of response bottleneck areas to generate new task scheduling policy constraints, namely load distribution constraints. These constraints are a set of limiting parameters that reflect the load status of each area in the current system. These include the maximum number of tasks assigned to a region per unit time, the minimum number of resources that can rotate, and the path selection entropy threshold. Based on these updated load distribution constraints, the task scheduling resource distribution is adjusted in real time to balance scheduling pressures across different areas and prevent the system from entering a local overload state.

[0111] After implementing the aforementioned scheduling logic adjustments, the warning identification module and the response scheduling module will execute subsequent warning response operations based on the updated scheduling strategy. The warning identification module analyzes real-time physiological parameter data using the updated path optimization rules and generates a second warning message. This information not only includes the anomaly type, occurrence time, and confidence score, but also selects the optimal warning generation unit as the execution entry point based on the new path optimization rules, improving the system's predictive capabilities and path selection rationality.

[0112] After receiving the second warning information, the response scheduling module first determines whether the current number of available response personnel is lower than the minimum optional threshold set by the system; if it is lower than the threshold, the task scheduling resources will be reconstructed according to the new load distribution constraints, such as expanding the available response circle, reducing the assignment density, etc.; if the number of optional personnel exceeds 1 person, the updated priority weight will be used to sort and select the optimal response personnel to ensure that the response task is assigned to the most timely and professional medical resources.

[0113] Through the above-mentioned multi-dimensional dynamic scheduling logic adjustment and feedback iteration mechanism, the present invention effectively improves the intelligent collaboration efficiency of the early warning task from identification to response, and overcomes the shortcomings of existing solutions in terms of regional overload, path bottlenecks and response imbalance.

[0114] Furthermore, the response scheduling module is also used to monitor the execution status of the response task. If the responder fails to complete the response within the specified time, the backup responder redistribution operation is triggered.

[0115] In this embodiment of the present invention, the response scheduling module not only intelligently dispatches response tasks but also continuously monitors and dynamically compensates for the execution status of assigned tasks. To prevent interruptions in the early warning response chain or intervention failures caused by delayed response personnel, uncompleted tasks, or unexpected unreachability, task status is tracked in real time after dispatch, and a backup response mechanism is introduced to ensure the integrity of the response loop.

[0116] Specifically, a limited response time window is set for each response task, and this time window can be dynamically adjusted based on the type and severity of the anomaly in the warning information. For example, for high-risk anomalies (such as cardiac arrest and tachypnea), the response time window can be set to less than 90 seconds; for routine anomalies (such as abnormal body temperature and mild blood pressure fluctuations), it can be set to 3 to 5 minutes. By integrating IoT positioning data, responder movement trajectories, and device interaction records, it is determined whether the target responder has completed arrival confirmation and task closure within the time window.

[0117] If no valid signs of mission completion are detected within the specified time, including but not limited to arrival confirmation punching, intervention equipment activation records, and mission feedback transmission, the response mission will be automatically determined to be at risk of abnormal delay or interruption. At this time, the response scheduling module triggers the backup response mechanism and re-screens candidate candidates from the backup response personnel based on the current mission priority, the scheduling pressure in the target area, and the status of the on-duty medical resource pool.

[0118] During the candidate screening process, priority is given to backup responders in the same physical partition, with a low current workload and appropriate intervention capabilities. A quick ranking is performed based on their task assignment priority and historical response performance, and the optimal responder is selected for redistribution. Once the backup responder confirms the task acceptance, the dispatch record is regenerated and synchronously written to the current response chain node to ensure the continuity of the task path.

[0119] During the above operation process, the present invention will also record the task response behavior of the original responder with labels, including fields such as failure to respond in time, response interruption, and backup trigger, and use it as a basis for negative adjustment of the weight factor in the next round of task allocation, thereby forming a closed-loop task assignment mechanism that integrates incentives and constraints.

[0120] By introducing dynamic identification of response status and backup response redistribution strategy based on timeliness monitoring, the present invention significantly improves the closed-loop success rate of the response chain, effectively reduces the risk of system response failure due to single-point delay, and provides strong technical support for high-timeliness response tasks in intensive care unit scenarios.

[0121] Example 2 A second aspect of the present invention discloses an early warning method for an intensive care unit based on the Internet of Things, the method comprising:

[0122] Collect the patient's physiological parameter data in real time;

[0123] Performing an abnormal state recognition operation based on the physiological parameter data through a pre-trained early warning model and outputting a first early warning message;

[0124] Determine the response task based on the first warning information, and determine the responder based on the response task, warning generation unit information, the current location of the medical staff, task load, professional matching and historical response efficiency score, and assign the response task to the responder;

[0125] Record the medical staff's response arrival, intervention type, and treatment results according to the preset cycle as response closed-loop data;

[0126] An early warning response responsibility chain is constructed based on the first early warning information within the cycle and the response closed-loop data, and a scheduling logic adjustment operation is performed based on the early warning response responsibility chain.

[0127] It should be noted that the specific implementation process of Example 2 is similar to that of Example 1 and will not be repeated in this embodiment.

[0128] Finally, it should be noted that the above-mentioned embodiments include multiple parallel implementation methods of the present invention, and deleting or otherwise adjusting one or more of the implementation methods will not affect the implementation of the solution. In addition, the early warning system and early warning method for intensive care units based on the Internet of Things disclosed in the embodiments of the present invention are only preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An early warning system for intensive care units based on the Internet of Things, characterized in that: The system further comprises: Vital signs collection module, used to collect patients' physiological parameter data in real time; an early warning identification module, configured to perform an abnormal state identification operation based on the physiological parameter data using a pre-trained early warning model and output a first early warning message; a response scheduling module, configured to determine a response task based on the first warning information, determine a response personnel based on the response task, warning generation unit information, the current location of the medical staff, task load, professional matching, and historical response efficiency score, and assign the response task to the response personnel; The response closed-loop recording module is used to record the response arrival status, intervention operation type and processing results of medical staff according to a preset period as response closed-loop data; The early warning feedback module is used to build an early warning response responsibility chain based on the first early warning information within the cycle and the response closed-loop data, and perform scheduling logic adjustment operations based on the early warning response responsibility chain.

2. The early warning system for intensive care units based on the Internet of Things according to claim 1, characterized in that: The early warning response responsibility chain includes node information and link relationships; wherein, The node information includes the early warning generation unit identifier, the dispatching unit identifier, the responder identifier, the intervention operation type and the intervention result status; The link relationship includes the delay relationship, task dependency and operation causal relationship between nodes.

3. The early warning system for intensive care units based on the Internet of Things according to claim 2, characterized in that: The process of building the early warning response responsibility chain includes: Conduct temporal correlation analysis on each warning event and reconstruct the response path from warning generation, task dispatch, response arrival, intervention execution to processing completion; Through the graph modeling mechanism, each response link is abstracted as a node according to the response path, and the operation timing, data dependency and causal relationship are mapped into edge relationships, generating a warning response responsibility chain structure diagram with response path, multi-stage delay parameters between nodes, intervention action information and task result identification.

4. The early warning system for intensive care units based on the Internet of Things according to claim 3, characterized in that: After constructing the early warning response responsibility chain structure diagram, high-risk nodes and response bottleneck areas in the early warning process are identified based on the early warning response responsibility chain structure diagram. The high-risk node identification process includes: A multi-dimensional performance evaluation index is constructed based on the average response delay, task completion rate and intervention effectiveness of each node in the early warning response responsibility chain structure diagram; A weighted calculation is performed on each node according to the multi-dimensional performance evaluation indicators to obtain the bottleneck risk score of each node. Nodes with bottleneck risk scores higher than the preset threshold are regarded as high-risk nodes.

5. The early warning system for intensive care units based on the Internet of Things according to claim 4, characterized in that: The process of identifying the response bottleneck area in the early warning process includes: Based on the adjacency relationship and path overlap of high-risk nodes in the early warning response responsibility chain structure diagram, areas with continuous high-risk nodes or high overlap of multiple early warning task paths are identified as response bottleneck areas.

6. The early warning system for intensive care units based on the Internet of Things according to claim 5, characterized in that: The execution scheduling logic adjustment operation specifically includes: Dynamically adjust the priority weights of medical staff's task allocation based on the type and spatial distribution of high-risk nodes; Adjust the path optimization rules in the early warning model based on bottleneck area feedback; Determine the response frequency and path coverage corresponding to the response bottleneck area, and update the load distribution constraints in the task scheduling process based on the response frequency and path coverage.

7. The early warning system for intensive care units based on the Internet of Things according to claim 6, characterized in that: The first warning information includes the abnormality type, abnormality occurrence time, confidence score and warning generation unit information; After executing the scheduling logic adjustment operation, the warning identification module and the response scheduling module are further used to perform a warning response based on the adjusted scheduling logic; wherein, The warning identification module performs a warning response based on the adjusted scheduling logic including: The warning model generates second warning information for the newly acquired physiological parameter data based on the adjusted path optimization rule; the warning generation unit information in the second warning information is determined based on the adjusted path optimization rule.

8. The early warning system for intensive care units based on the Internet of Things according to claim 7, characterized in that: The response scheduling module performs an early warning response based on the adjusted scheduling logic, including: When the number of determined responding personnel is lower than the preset minimum selectable personnel threshold, the current task scheduling resources are adjusted according to the updated load distribution constraints; When the number of identified responders is more than one, the target responders are determined based on the adjusted priority weights of the medical staff's tasks.

9. The early warning system for intensive care units based on the Internet of Things according to any one of claims 1 to 8, characterized in that: The response scheduling module is also used to monitor the execution status of the response task. If the responder fails to complete the response within the specified time, the backup responder redistribution operation is triggered.

10. An early warning method for an intensive care unit based on the Internet of Things, the method being implemented by the system according to any one of claims 1 to 9, characterized in that: The method comprises: Collect the patient's physiological parameter data in real time; Performing an abnormal state recognition operation based on the physiological parameter data through a pre-trained early warning model and outputting a first early warning message; Determine the response task based on the first warning information, and determine the responder based on the response task, warning generation unit information, the current location of the medical staff, task load, professional matching and historical response efficiency score, and assign the response task to the responder; Record the medical staff's response arrival, intervention type, and treatment results according to the preset cycle as response closed-loop data; An early warning response responsibility chain is constructed based on the first early warning information within the cycle and the response closed-loop data, and a scheduling logic adjustment operation is performed based on the early warning response responsibility chain.

Images